Effects of promoting drinking guidelines on alcohol consumption behaviours and behavioural influences

Parts of this chapter have been reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/.

Introduction

There is little robust evidence available on if and how the promotion of drinking guidelines affects alcohol-related outcomes.13,28 One reason for this is that health authorities often do not actively promote drinking guidelines after publishing them. Our analysis of promotional activity (see Chapter 2) and news media coverage (see Chapter 3) suggests that this was the case for the 2016 revisions to the UK guidelines. We identified no large-scale promotional activity by the UK Government, PHE or any third-sector organisation. We also identified no large peaks in news media coverage after the initial announcement of the revised guidelines in January 2016.

The lack of promotional activity has three implications for our evaluation. First, it indicates that we should focus on January 2016 as our intervention point. Although the government did not formally adopt the revised guidelines until it had completed a public consultation in August 2016, there was no substantial change in news reporting at this time, and this leaves the January announcement as the only large-scale promotional event. Second, the lack of any large-scale promotional activity after January 2016 means that we cannot undertake our planned analysis of the effect of promotional campaigns that occurred at different points in time and we treat January 2016 as the sole intervention point. Third, our preregistered protocol and earlier chapters of this report state that our aim is to evaluate the impact of promoting revised drinking guidelines on alcohol consumption behaviour. However, in the absence of significant promotional activity, it is more accurate to describe the analyses below as an evaluation of the impact of announcing such guidelines, and we will use this language henceforth.

We planned to use a monthly repeat cross-sectional survey of adults in England to conduct a series of analyses to evaluate the impact of announcing the guidelines in three stages. First, we describe trends in respondents’ awareness and knowledge of drinking guidelines, where they report being exposed to the guidelines and influences on change in drinking behaviours. As part of this, we test whether or not the average level of measures of these outcomes differed between the pre- and post-intervention time point (i.e. January 2016) and across key population subgroups defined by sociodemographics and alcohol consumption level. Second, the primary analysis uses segmented regression to test whether or not alcohol consumption behaviour changed following the announcement of the revised drinking guidelines. Third, we assess the robustness of our results and competing interpretations via a series of secondary analyses involving alternative outcome measures and model specifications.

Methods

Data

The analyses use data from the Alcohol Toolkit Study (ATS). The ATS methods are summarised here and are described in full elsewhere.144 The study is nested within a wider omnibus survey conducted by the research agency Ipsos MORI (Ipsos MORI, London, UK) and collects data from nationally representative samples of approximately 1700 adults aged ≥ 16 years living in private households in England each month. It samples participants using a hybrid between random location sampling and quota sampling, in which England is split into 171,356 areas containing approximately 300 households and a random sample is drawn from strata defined by area-level geographic and sociodemographic profiles. Each area is allocated to an interviewer, who conducts in-home computer-assisted interviews until they fill quotas tailored to each area and based on factors that influence the probability of an individual being at home (e.g. employment status, age and sex).

Although the ATS has collected data since March 2014, this project allowed additional questions relating to drinking guidelines to be added to the survey between November 2015 and October 2017. Therefore, the primary evaluation analysis used data collected between March 2014 and October 2017 (i.e. 22 months before and 22 months after the revised guidelines were announced) and descriptive or secondary analyses use data collected between November 2015 and October 2017 (i.e. 2 months before and 22 months after the announcement).

Measures for descriptive analyses

The ATS questionnaire asks drinkers about their awareness and knowledge of and exposure to the guidelines in sequence, as described below. We define drinkers as those who did not respond ‘never’ to the first question of the Alcohol Use Disorders Identification Test – Consumption (AUDIT-C) (see Primary outcome measures). Only drinkers who are aware of drinking guidelines are asked about their knowledge and only those who give a numerical response to the knowledge questions are asked about their exposure. Thus, the baseline sample is progressively smaller across each of these questions. The full Alcohol Use Disorders Identification Test (AUDIT) questionnaire is provided as an additional file (www.journalslibrary.nihr.ac.uk/programmes/phr/156301/#/documentation; accessed April 2020).

Awareness

The ATS measures awareness of guidelines using the question: ‘Before this interview, had you ever heard of there being a recommended maximum number of alcohol units people should drink in a day or a week? This is sometimes known as a “drinking guideline”’.

Knowledge

The questionnaire measured knowledge using the question ‘How many units per day or per week is the drinking guideline for males/females?’. Participants were asked only about their own sex’s guideline, with those identifying with neither sex excluded from analyses as it was unclear which guideline was appropriate to them. Participants could respond in either units per week or units per day to allow the question to remain unchanged before and after the revised guidelines were announced. This means that women giving the 1995 daily guideline would still be classed as having given the correct answer if they said 2 units per day rather than 14 units per week, but not if they said 1 unit per day. We trichotimise responses as more than, exactly or below 14 units per week or 2 units per day. This allows us to examine whether or not announcing the new guidelines increased the proportion of people saying the guideline was 14 units per week (or the daily equivalent).

Exposure

Interviewers provided participants with a list of places and scenarios and asked if they had seen, read or heard about the guideline figure they provided in each of them during the last month. The full list of options is as follows: TV/radio, newspapers/magazines, websites/social media, shops/supermarkets, pubs/bars/restaurants, at place of work/study, talking to health professional, posters/booklets at health service, talking to friends/family/colleagues, other, and none of the above. Responses were dichotomised as yes or no and there was no limit on the number of places participants could say ‘yes’ to.

Capability, opportunity, motivation and behaviour determinants of behaviour change

The ATS included an additional set of 10 questions with ordinal responses scales to capture the capability, opportunity and motivation components of the capability, opportunity, motivation and behaviour (COM-B) model of behaviour change.68 COM-B is a well-supported model that integrates core constructs present in many earlier theories. It specifies that capability, opportunity and motivation interact within a system to generate behaviour. The 10 questions are presented in Table 4. The questions were designed for this project and were influenced by questions found to predict smoking behaviour in the Smoking Toolkit Study.145 Responses to the COM-B questions are dichotomised into ‘positive’ and ‘negative’ groupings, as shown in Table 4. The first question is an exception as it asks how many units of alcohol can be regularly consumed in a day without significantly harming health, and the first two responses (1 and 2 units) are classed as positive to be consistent with the revised guidelines.

Sociodemographic measures

Ipsos MORI collects a battery of sociodemographic measures from respondents as part of the wider omnibus survey.144 Of these, we use participants’ sex, age (16–34, 35–64 and ≥ 65 years) and social grade, an occupation-based measure of socioeconomic status derived from the National Readership Survey.147 Respondents are classified as grade AB (high), C1C2 (intermediate) or DE (low). The descriptive analyses also explore differences by alcohol consumption level using the graduated frequency measure of alcohol consumption described in Secondary outcome measures. Respondents are classified into three groups based on their weekly alcohol consumption: moderate drinkers, consuming < 14 units per week; increasing-risk drinkers, consuming between 14 and 35 units per week (women) or between 14 and 50 units per week (men); and high-risk drinkers, consuming > 35 units per week for women and > 50 units per week for men.148

Measures for primary and secondary evaluation analyses

Primary outcome measures

The primary outcome is alcohol consumption behaviour, which the ATS measures using participants’ AUDIT-C scores. AUDIT-C is the short form of the AUDIT and is a validated screening test for heavy drinking and/or active alcohol abuse or dependence and has demonstrated excellent reliability and responsiveness to short-term change.149,150 Although the ATS collects data using the extended AUDIT-C, which provides additional response options on one question, we collapsed these additional options to match those in the standard questionnaire before calculating participants’ scores. The AUDIT-C questionnaire comprises three questions that ask about respondents’ alcohol consumption frequency (never, monthly or less, 2–4 times per month, 2 or 3 times per week or ≥ 4 times per week), quantity of standard drinks consumed on a typical drinking day (1–2, 3–4, 5–6, 7–9, 10–12, 13–15 or ≥ 16) and frequency of drinking six or more standard drinks (never, monthly or less, monthly, weekly, or daily or almost daily). A standard drink is defined for ATS respondents as 1 UK unit (i.e. 8 g/10 ml ethanol). Each response option is scored from 0 to 4, with the extended AUDIT-C response options of ≥ 10 for the quantity question all scoring 4 points. Scores are then summed to give the total AUDIT-C score.

Secondary outcome measures

We also explore a series of secondary outcome measures to test the robustness of our results.

Alcohol consumption measured using graduated frequency

Between November 2015 and October 2017, the ATS also collected alcohol consumption data using a graduated frequency approach.151,152 This asks drinkers (1) on how many days they drank during the last 4 weeks, (2) how many units they consumed on their heaviest drinking day, (3) on how many days they drank at that level and (4) on how many days they drank at the following progressively decreasing levels starting from their maximum (31–40 units, 21–30 units, 16–20 units, 11–15 units, 8–10 units, 5–7 units, 3–4 units, 1–2 units). The number of days consuming each quantity was then multiplied by that quantity (with midpoints used for ranges) and the sum of the multiples was divided by 4 to give a measure of average weekly units consumed.

Alcohol consumption measured using alcohol duty data

We also use an aggregate-level measure of alcohol consumption that can be derived from the monthly alcohol duty bulletins released by Her Majesty’s Revenue and Customs (HMRC). This measure is litres of pure alcohol released for sale per adult aged ≥ 16 years. It is not subject to self-report biases but is subject to other general biases that are discussed by Meier et al.153 It may also be subject to temporal biases when used as a monthly time series, as it is here, to the extent that alcohol producers stockpile or release excess volumes of product onto the market for commercial reasons rather than to meet demand.

Hospital admissions for alcohol-related harm

Data on monthly hospital admissions for alcohol poisoning [International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10): T51.0, T51.1, T51.9] and assaults (ICD-10: X85–Y09) provide objective proxy measures of change in heavy episodic drinking and alcohol-related violence, respectively. We use raw counts of monthly hospital admission for each of two groups of ICD-10 codes.

Covariates

The evaluation analyses control for change in alcohol prices and weather. We control for prices using monthly retail price indices for four beverage categories (on-trade beer, off-trade beer, on-trade wine and spirits, and off-trade wine and spirits) from the ONS.154 The Met Office provides freely accessible monthly temperature data for locations in England but not for the country as a whole.155 Therefore, we use the Met Office’s Hadley Centre Central England Temperature (HadCET), which records the average monthly temperature across an area of central England approximately enclosed by London, Bristol and Lancashire.156

Analysis

Descriptive analyses

All descriptive analyses are based on data that are weighted using an iterative sequence of weighting adjustments that compares variables against separate nationally representative target profiles for sex, working status, children in the household, age, social grade and region until the targets are met. Comparisons of outcome measures pre and post intervention are based on the average response of all respondents surveyed in each of the two time periods and the significance of differences is assessed using chi-squared tests. Comparisons between population subgroups defined by sociodemographic characteristics or alcohol consumption levels are based on post-intervention data only so as to assess whether or not differences are present between groups after the guidelines are announced. The significance of differences is tested using multivariate logistic regressions adjusting for each of the other subgroup characteristics.

Primary analyses (planned): alcohol consumption

We used an interrupted time series design to assess the effect of the drinking guidelines on AUDIT-C scores using individual-level data. Data were analysed with segmented regression using generalised additive models (GAM) [R (The R Foundation for Statistical Computing, Vienna, Austria) package mgvc], which allow the analyst to account for complex seasonality. The analyses use the Gaussian family with identity link. Each GAM modelled the trend in the dependent variable in the pre-intervention period, any immediate step change in the dependent variable when the guidelines were introduced and any change in the trend in the post-intervention period relative to the pre-intervention period. We ran unadjusted models and models adjusting for temperature and monthly price index for four beverage categories. All models were adjusted for seasonality.

Secondary analyses (planned)

We evaluated the impact of announcing the revised drinking guidelines on two alternative alcohol consumption measures (i.e. graduated frequency and alcohol duty data). The ATS included graduated frequency questions only from November 2015, 2 months before the new drinking guidelines came into effect. As the series was not long enough to model the pre-intervention trend, the analysis could not use the preplanned interrupted time series design. Instead, we conducted a simple pre/post analysis. We ran unadjusted and adjusted GAMs, controlling for alcohol prices and temperature, at the individual level for graduated frequency and at the aggregate level for sales data. The partial autocorrelation function indicated the presence of first-order autoregression for the sales data. This was confirmed by comparing models with various fitted autoregression moving average (MA) terms using the Akaike information criterion (AIC). This autocorrelation was taken into account by extending the GAM to a generalised additive mixed model (GAMM) using the Gaussian family with identity link.

We also use an interrupted time series design to evaluate the impact of announcing the revised guidelines on incidence of alcohol poisoning and assaults. As above, the partial autocorrelation function indicated the presence of first-order autoregression. We again confirmed this by comparing models with various fitted autoregression and MA terms using the AIC and then took this autocorrelation into consideration by extending the GAM to a GAMM. The hospital admissions data were counts and overdispersed, which required negative binomial regression models. As above, each GAM modelled the trend in the dependent variable in the pre-intervention period, any immediate step change in the dependent variable when the guidelines were introduced and any change in the trends in the post-intervention period relative to the pre-intervention period. We again ran unadjusted models and adjusted models.

In addition to using the January 2016 announcement as the break point in the segmented regression models, we also used an iterative procedure to identify statistically if there were any alternative significant break points present in the AUDIT-C data. This analysis was run using the R segmented package, which compares models specified with different break points. The analyst must give a starting value for the break point parameters, but, as the package implements bootstraps, restarting it makes the algorithm insensitive to starting values.157 After identifying an alternative break point, we then re-ran the GAM to model the trend before the identified break point and the trend following the identified break point.

We also tested for a pulse effect (i.e. an immediate but short-term change) following the intervention by running GAMs with a 2-month (January to February 2016) and 3-month (January to March 2016) pulse. Furthermore, we tested whether a quadratic or cubic model fitted the trends in the data better than the linear model in our primary analyses by running segmented polynomial GAM regression models.

Sensitivity analyses (unplanned)

Variability in AUDIT-C post intervention was greater than anticipated. Such variability reduces power to detect an effect and can produce poor model specification if there are significant outliers or influential points. To mitigate this variability, sensitivity analyses were included that specified a longer post-intervention period, until February 2018.

Analyses not undertaken

We did not proceed with the planned difference-in-difference analyses to test whether or not effects were larger in drinkers exposed to the guideline and a pathway analysis to understand the relationship between changes in the COM-B measures after assessing the final results from the primary analysis.

Results

Descriptive analyses: awareness, knowledge and exposure

Table 5 summarises the characteristics of the sample and provides pre- and post-intervention data on the primary and secondary outcome measures, as well as measures of statistical differences in measures between these time points. Figure 7 shows that the proportion of respondents who were drinkers varied across the study period, between 72.8% in April 2015 and 62.7% in November 2016. The proportion of respondents who were drinkers was higher in the pre-intervention period than in the post-intervention period (70.4% vs. 65.7%, χ² = 171.5; p < 0.001). During the shorter period from November 2015, when the ATS included questions related to drinking guidelines, the proportion of respondents who were drinkers peaked at 70.3% in November 2015.

FIGURE 7 The proportion of respondents who are (a) drinkers; (b) drinkers who are aware of the guidelines; and (c) drinkers who can correctly identify the guideline. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

Figure 7 also shows that awareness of guidelines among drinkers was high across the shorter study period, at an average of 86.9%, and did not change substantively after the revised guidelines were announced. Furthermore, Figure 7 shows that knowledge of guidelines did change to a modest degree during the study period. Among drinkers who were aware of guidelines, in December 2015, 20.6% said that the guideline was 14 units per week (or 2 units per day), and this rose to 29.0% in January 2016, when the guidelines were announced. However, this increase in knowledge dissipated over subsequent months, and only 24.1% of drinkers identified the guideline correctly in October 2017.

Overall, the proportion of drinkers who correctly identified the guideline was slightly lower in the pre-intervention than the post-intervention period, suggesting a small improvement in knowledge (20.6% vs. 23.7%, χ² = 4.2; p = 0.04). The analyses show a similar pattern of results if those who said that the guideline was < 14 units are combined with those who gave the correct answer.

The guideline for men changed to a greater degree than the guideline for women, so we compared the change in knowledge between the two sexes (Figure 8). Among drinkers, in December 2015, 9.1% of men who were aware of the guidelines said that the guideline was 14 units per week, and this rose to 23.8% in January 2016 before falling to 20.1% in October 2017. Among women, the change was smaller: the proportion of drinkers who stated that the guideline was 14 units per week was 33.2% in December 2015, rising to 34.6% in January 2016 and falling to 28.6% in October 2017. The proportion of male drinkers identifying the guideline correctly was higher in the post-intervention period than in the pre-intervention period (17.6% vs. 9.1%, χ² = 42.9; p < 0.001), but this was not the case for women (30.1% vs. 31.7%, χ² = 0.9; p = 0.35).

FIGURE 8 The proportion of drinkers correctly identifying the guidelines, by sex. (Figure not reproduced here; see the original article.)

Exposure questions were asked only to the 68.8% of drinkers who were able to name a figure for the guideline consumption level irrespective of whether or not it was correct. There were only three locations in which more than 20% of this subsample reported exposure to the guideline in the last month (Figure 9). These locations were on the TV or radio, on product labels, and in newspapers or magazines. TV and radio was the most commonly cited source of exposure to the guideline, with 32.5% reporting last-month exposure in December 2015, rising to 64.8% following the guideline announcement in January 2016, which attracted prominent news coverage. Exposure via TV and radio then fell gradually to vary between 28.8% and 35.6% across 2017. Exposure via newspapers and magazines showed a similar peak and then decline but varied only between 12.6% and 24.3% across the study period. As would be expected, exposure via product labels did not peak when the revised guidelines were announced as product labels were not updated immediately and exposure via labels varied between 14.3% and 27.7% across the study period. Overall, slightly more of this subsample of drinkers reported exposure to their reported guideline figure post intervention than pre intervention (76.5% vs. 70.8%, χ² = 20.9; p < 0.001). However, a significant increase in exposure was found only in the case of TV and radio (39.1% vs. 34.2%). The findings on exposure did not differ substantively when limiting the subsample to those respondents who correctly identified the guideline rather than those identifying any guideline. Figure 10 shows that, across the study period, only between 20% and 30% of the subsample were not exposed to guidelines in any location during the last month.

FIGURE 9 The proportion of drinkers giving a figure for the guidelines who reported exposure to that figure in different locations during the last month. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

FIGURE 10 The proportion of drinkers giving a figure for the guidelines who reported no exposure to that figure in any location during the last month. (Figure not reproduced here; see the original article.)

There were differences between sociodemographic and consumption groups in likelihood of being a drinker as well as awareness, knowledge and exposure to the guideline (Figure 11; see Appendix 3 for full regression results). Respondents who were men, older or of a higher social grade were more likely to be drinkers. In general, awareness, knowledge and exposure were higher among those in higher social grades and those drinking at increasing- or high-risk levels. For example, awareness and knowledge varied particularly by social grade. Awareness fell from 92.4% among those in grade AB to 78.5% among those in grade DE, and 27.3% of those in grade AB identified the guideline correctly, compared with 18.1% of those in grade DE. Exposure showed more variation in social patterning depending on the medium, although rates of exposure were generally low in all cases and the most likely means of exposure for all groups were via product labels, TV or radio, and newspapers or magazines. Overall, those aged 16–24 years and heavier drinkers were more likely to report exposure in at least one medium. Older drinkers were more likely to report exposure through TV, radio, newspapers and magazines but younger drinkers were more likely to report exposure via product labels, licensed venues or talking to friends, family and colleagues. Those in lower social grades were less likely to report exposure via newspapers and magazines. Notably, heavier drinkers were more likely to report exposure by talking to health professionals, or from posters or booklets at health services, as well as by talking to friends, family or colleagues.

FIGURE 11 The proportion of respondents who are drinkers, aware of guidelines, know the correct guideline or have been exposed to the guidelines in different locations, by sociodemographic and consumption group. (a) Drinker; (b) aware of guidelines; (c) know correct guideline; (d) exposed via product labels; (e) exposed via TV or radio; (f) exposed via newspapers or magazines; (g) exposed via websites or social media; (h) exposed via shops or supermarkets; (i) exposed via pubs, bars or nightclubs; (j) exposed via place of work or study; (k) exposed via health professionals; (l) exposed via health services; (m) exposed via friends, etc.; (n) exposed via other means; and (o) exposed in no places. Light-blue bars indicate a significant difference compared with the left-most category within characteristic in multivariate logistic regressions adjusting for all listed characteristics. (Figure not reproduced here; see the original article.)

Descriptive analyses: capability, opportunity and motivation

Figures 12–15 show trends in the dichotomised COM-B measures over the study period. Although some of these measures did change significantly after the announcement of revised guidelines, these changes were small and there is little evidence of any substantial change over time.

FIGURE 12 The proportion of respondents answering positively to capability-related outcome measures over time. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

FIGURE 13 The proportion of respondents answering positively to opportunity-related outcome measures over time. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

FIGURE 14 The proportion of respondents answering positively to automatic motivation-related outcome measures over time. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

FIGURE 15 The proportion of respondents answering positively to reflective motivation-related outcome measures over time. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

Capability

Across the whole study period, 35.5% of drinkers thought that they could regularly drink no more than 2 units per day before doing significant harm to their health. This did not change significantly after the announcement of revised guidelines (34.25% pre change vs. 35.6% post change, χ² = 1.26; p = 0.26). Unlike the knowledge question above, there was also no change among male drinkers (24.8% pre change vs. 26.9% post change, χ² = 1.67; p = 0.20). Most drinkers indicated that they would find it easy to drink ≤ 3 units per day, with 82.0% saying this across the study period, although only 26.0% indicated that they often tracked their units. In both cases, there was no significant change between the pre- and post-intervention periods.

Opportunity

The proportion of drinkers who indicated that their lifestyle made it easy to drink ≤ 3 units per day was 79.2% across the study period and increased slightly after the announcement of revised guidelines (76.9% pre change vs. 79.4% post change, χ² = 6.5; p = 0.01). The proportion who knew where to get advice to cut down drinking was also high across the study period, at 82.0%, and, again, increased slightly post intervention (78.6% pre change vs. 82.3% post change, χ² = 16.0; p < 0.001).

Motivation

Only a minority of drinkers (23.9%) were concerned about drinking more alcohol than was good for them, but 51.0% of drinkers wanted to avoid consuming this amount, although 76.4% wanted to keep their drinking within safe limits. Only the proportion wanting to avoid drinking more than was good for them increased post intervention, and the increase was small (48.7% vs. 51.2%, χ² = 3.9; p = 0.05). A large minority, comprising 39.6% of drinkers, were trying not to drink more than was good for them across the study period, and 77.4% indicated that they intended to drink within safe limits. Neither of these two measures changed post intervention.

Figure 16 shows that there are several significant differences in the COM-B measures between sociodemographic and, particularly, drinking groups during the post-intervention period (see Appendix 3 for full regression results).

FIGURE 16 The proportion of respondents giving positive responses to COM-B questions by sociodemographic and consumption. (a) Knowledge: 1–2 units is most can consume on a day before it does significant harm to health; (b) perceived capability: easy to drink ≤ 3 units per day; (c) skills: often keeps track of units; (d) social opportunity: lifestyle makes it easy to drink ≤ 3 units per day; (e) social opportunity: knows where to get advice on cutting down on drinking; (f) automatic motivation: concerned about drinking more alcohol than is good for you; (g) automatic motivation: wants to avoid drinking more than is good for you; (h) reflective motivation: trying to avoid drinking more than is good for you; (i) automatic motivation: wants to keep drinking within safe limits; and (j) reflective motivation: intends to drink within safe limits. Light-blue bars indicate a significant difference compared with the left-most category within characteristic in multivariate logistic regressions adjusting for all listed characteristics. (Figure not reproduced here; see the original article.)

Although the descriptive analyses above show that men’s knowledge of the correct guideline increased over the study period, women were significantly more likely than men to say that ≤ 2 units per day is the most they can consume regularly before it harms their health. Differences on other measures, although statistically significant in most cases, were smaller. The exceptions are that women were more likely than men to both want and intend to keep their drinking within safe limits.

With regard to age, the largest differences were that those aged ≥ 65 years were more likely to say that they would find it easy to drink ≤ 3 units per day and that their lifestyle made this easy. This age group was also more likely to want to, and intend to, keep their drinking within safe limits. Differences between those aged 16–34 years and those aged 35–64 years were generally small, although often statistically significant. The largest difference was that those aged 16–34 years were less likely to say that their lifestyle made it easy to drink < 3 units per day.

Differences by social grade were of similar size to those by age. Those in social grade AB were more likely than those in lower social grades, and particularly those in grade DE, to say that they would find it easy to drink ≤ 3 units per day, that their lifestyle made it easy to do so and that they knew where to get advice on cutting down drinking. They were also more likely to say that they wanted to avoid drinking more than is good for them and both wanted and intended to stay within safe limits.

The differences between drinker groups were much larger than the sociodemographic differences. Although the descriptive analyses above show that increasing- and high-risk drinkers are more likely to know and be exposed to the guidelines, the COM-B measures suggest that they are much less likely to say that ≤ 2 units per day is the most they can consume regularly before it harms their health. Increasing-risk and especially high-risk drinkers are also much less likely to say that they are able to drink ≤ 3 units per day or that their lifestyle would make this easy. Similarly, increasing-risk and especially high-risk drinkers were much less likely to say that they wanted to or were trying to avoid drinking more than is good for them or that they wanted to or intended to keep their drinking within safe limits.

Primary analyses (planned): alcohol consumption

The primary evaluation analyses showed that there was no significant change in the level of AUDIT-C scores in the unadjusted model after the announcement of the revised guideline, but there was a significant change in the AUDIT-C score trend (Figure 17 and Table 6). In December 2015, the modelled average AUDIT-C score was 2.719, and it was decreasing by 0.003 each month. After the January 2016 announcement, AUDIT-C scores increased immediately but non-significantly to 2.720 [β = 0.001, 95% confidence interval (CI) –0.079 to 0.099] and the trend changed significantly such that scores subsequently increased by 0.005 each month (β = 0.008, 95% CI 0.001 to 0.015). This is equivalent to an additional 0.5% of the population increasing their AUDIT-C score by 1 point each month by, for example, drinking four or more times per week rather than two or three times per week. The results did not change after adjusting for average temperature, but the change in trend was no longer significant when controlling for alcohol prices or in the fully adjusted model.

FIGURE 17 Primary analysis showing AUDIT-C scores before and after the announcement of revised drinking guidelines in January 2016. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

Secondary analyses (planned)

Tables 7 and 8 show the results of analyses using alternative graduated frequency and taxation measures, respectively (also shown in Figures 18 and 19). There was a non-significant increase in alcohol consumption measured by graduated frequency of 0.1 units after the announcement of revised guidelines (β = 0.105, 95% CI –0.652 to 0.862). Similarly, there was an immediate non-significant increase in per capita litres of alcohol released for sale of 0.006 l (β = 0.006, 95% CI –0.025 to 0.036), and the trend in this measure changed non-significantly from declining by 0.001 l per capita per month to remaining stable (β = 0.001, 95% CI –0.001 to 0.004).

FIGURE 18 Mean weekly consumption in units measured by graduated frequency before and after the announcement of revised drinking guidelines in January 2016 (data only available from November 2015). (Figure not reproduced here; see the original article.)

FIGURE 19 Litres of ethanol released for sale per capita adult before and after the announcement of revised drinking guidelines in January 2016. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

Tables 9 and 10 show the results for measures of alcohol-related hospital admissions for assaults and alcohol poisonings, respectively (also shown in Figures 20 and 21). There was a significant step change and change in trend for incidence of assault hospitalisations after the announcement of the revised guidelines. Hospitalisations for assault decreased by 7.3% immediately after the announcement [incidence rate ratio (IRR) 0.927, 95% CI 0.875 to 0.983], but this was partly counteracted by a change from a flat trend pre intervention to a 0.5% increase in the incidence of assault hospitalisations each month (IRR 1.005, 95% CI 1.001 to 1.009). The step change but not the change in trend remained in the adjusted models. There was also a significant step change of –15.4% in the incidence of alcohol poisoning hospitalisations (IRR 0.846, 95% CI 0.735 to 0.975), whereas the trend changed non-significantly from increasing by 0.9% each month to increasing by 0.5% each month (IRR 0.996, 95% CI 0.984 to 1.007). The step change was no longer significant after adjusting for the temperature trend.

FIGURE 20 Number of hospitalisations for assaults before and after the announcement of revised drinking guidelines in January 2016. (Figure not reproduced here; see the original article.)

FIGURE 21 Number of hospitalisations for alcohol poisoning before and after the announcement of revised drinking guidelines in January 2016. (Figure not reproduced here; see the original article.)

We used an iterative procedure to identify any statistically significant break points in the AUDIT-C score trend. Table 11 and Figure 22 show that the procedure identified a break point in June 2015, 7 months before the announcement of the revised guidelines. In this model, AUDIT-C scores declined by 0.005 points per month up to June 2015 and began to increase significantly, by 0.011 points per month, thereafter (β = 0.011, 95% CI 0.003 to 0.019).

FIGURE 22 Model-defined break point in trend in mean AUDIT-C score across the study period. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

We also tested for the presence of pulse effects at 2 and 3 months after the announcement of the revised guidelines. Table 12 shows that we found evidence in both cases of a significant pulse effect, such that AUDIT-C scores were significantly lower, by 0.184 points for 2 months after the intervention (β = –0.184, 95% CI –0.292 to 0.076) and by 0.126 points for 3 months after the intervention (β = –0.126, 95% CI –0.218 to 0.034). These effects are equivalent to 1.8% of the population reducing their AUDIT-C score by 1 point for 2 months or 1.3% of the population reducing their AUDIT-C score by 1 point for 3 months. We also ran unplanned sensitivity analyses to assess when the identified pulse effects dissipated. The results in Table 13 suggest that the pulse effects lasted for a 4-month period after the announcement of the revised guidelines. Figure 23 shows the pulse effect visually.

FIGURE 23 Short-term or pulse effect on AUDIT-C following the announcement of revised drinking guidelines in January 2016. Reproduced with permission from Holmes et al.143 This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/. (Figure not reproduced here; see the original article.)

The AUDIT-C scores showed greater than anticipated variation across the post-intervention period, so we ran unplanned sensitivity analyses to assess whether or not the results changed when using an extended post-intervention period of 48 months (i.e. until February 2018). Table 14 shows that we found similar results to the primary analysis. The pre-intervention trend of a reduction in AUDIT-C scores of 0.002 units each month changed significantly to an increase in AUDIT-C scores of 0.010 units each month (β = 0.010, 95% CI 0.004 to 0.016). Furthermore, in planned sensitivity analyses, we tested whether quadratic or cubic segmented regressions models provided a better fit.

Discussion

The analyses above suggest that announcing revised drinking guidelines did not lead to a step-change decline or a reduction in the longer-term trend in self-reported alcohol consumption behaviour. Further analyses inconsistently indicated small and short-term reductions in various secondary outcomes relating to alcohol consumption. There were small and selective impacts on drinkers’ awareness and knowledge of the guidelines and key influences on behaviour change. We summarise the results below and then discuss the key strengths and limitations of the analysis. Chapter 5 presents a broader summary of the project findings, drawing on previous chapters, as well as a discussion of the implications for policy, practice and future research of the main evaluation analyses presented here.

Summary of findings

The evaluation analyses focused on alcohol consumption behaviour and showed mixed effects of announcing revised guidelines depending on the outcome measure and analytical approach used. The primary analysis suggested that the announcement of the revised drinking guidelines coincided with a change in trend towards slightly higher AUDIT-C scores. However, this change in trend was not significant in some adjusted models and secondary analyses showed that it was not present in analyses using alternative alcohol consumption measures. Furthermore, secondary analyses also suggested that any change in alcohol consumption behaviour was small and short term (i.e. a pulse effect) and any change in trend may have begun 6 months before the revised guidelines were announced. In contrast, the analyses of hospital admission data did suggest that assault admissions fell by 7.2% and alcohol poisoning admissions fell by 15.4% immediately after the guideline announcement. These results should, however, be interpreted with caution as they contradict findings for both self-reported and objective measures of alcohol consumption, they are larger than expected given the modest change in guidelines and lack of promotional activity, and they do not align well with the descriptive analyses discussed below. It appears more likely that the reductions in admissions are attributable to other factors or an artefact of the analyses, and we discuss this further in Strengths and limitations. Overall, it appears unlikely that the guideline announcement led to a change in alcohol consumption behaviour. Any change that did occur was not sufficiently substantial or sustained to be identified independent of other overlapping changes.

The descriptive analyses provide useful insights into key factors affecting drinkers’ ability to make use of the guidelines and to change their drinking behaviour. Awareness of drinking guidelines is generally high, a small majority of drinkers are exposed to guidelines each month, and most report finding it easy to consume only moderate amounts. However, only around one-fifth of drinkers can correctly identify the guideline, few keep track of the units they consume and most drinkers also think that they can regularly exceed the guideline without doing significant harm to their health. Motivations related to alcohol consumption also vary depending on the questions asked, with drinkers generally wanting and intending to drink within safe limits but less likely to say that they are trying to avoid drinking more than is good for them or want to do so. These results suggest that, although the concept and communication of guidelines are well established, albeit by passive means rather than active promotional campaigns, drinking guidelines are not achieving the government’s aim to:

Support people to make informed choices about healthier and responsible drinking.

They are also unlikely to function effectively as a behaviour change intervention.

Of particular concern are differences in measures relating to guidelines across sociodemographic groups, as these may give rise to or perpetuate alcohol-related health inequalities.158 Drinkers in social grade DE are of particular concern because those of lower socioeconomic status are at greater risk from their alcohol consumption.159–161 It is therefore worrying that the descriptive analyses point to substantial deficits in communication of the guidelines and greater impediments to behaviour change within this group. Drinkers in social grade DE are less likely than those in other social grades to be aware of guidelines or to know what the guideline is. They also find it more difficult to drink at moderate levels, are less motivated to do so and are less sure where to get advice on cutting down. Smokers of lower socioeconomic status, compared with those of higher socioeconomic status, also find quitting smoking more difficult but, in contrast, tend to be at least as educated on harms, as motivated to stop and as willing to use support.162,163 There were fewer and smaller differences between other sociodemographic groups. Only the guideline for men changed substantially, and men were more likely to say that their guideline was 14 units per week (or 2 units per day) after January 2016. However, they remained substantially less likely than women to know this guideline, and knowledge remained low among both sexes. Drinkers aged 16–24 years are more likely to be exposed to guidelines, particularly via product labels, but find it harder to drink at moderate levels and are less motivated to do so.

Those drinking at increasing-risk and, particularly, high-risk levels are also of concern as it these groups that the government particularly wishes to make healthier and responsible choices about drinking (paraphrased).5

In line with previous studies of alcohol-related health information,50,52 heavier drinkers are more likely than those drinking at moderate levels to be aware of and exposed to the guideline, although they are not more likely to correctly identify it. Heavier drinkers are also much less likely to say that 2 units a day is the most they can drink without harming their health, suggesting that knowledge of the guideline is not translating into acceptance of it. Furthermore, heavier drinkers find it harder to drink at moderate levels, are less motivated to do so and are less likely to track the units they consume.

Strengths and limitations

To the best of our knowledge, this is the most detailed study to date of the impact of announcing or promoting drinking guidelines. Strengths include the use of a monthly time series of nationally representative survey data over a time period varying between 24 and 44 months, depending on the analysis. We analyse a range of outcome measures that align with concepts addressed in previous evaluations of drinking guidelines and alcohol-related health warnings (e.g. awareness, knowledge and exposure), the behaviour change literature (e.g. the COM-B measures) and key public health outcomes (e.g. alcohol consumption and hospital admissions). These measures allow us to test the robustness of our results and to interpret and contextualise the findings of the primary analysis within a broader understanding of drinkers’ interactions with the guidelines and their perspectives on alcohol and associated health risks. The detailed data series also permit robust statistical analyses that control for important confounders. A particular strength is our exploration of alternative temporal processes (including different break points or pulse effects and non-linear polynomial trends) that point towards different interpretations of the primary analysis.

There are, however, a number of important limitations. First, our primary outcome measure of alcohol consumption behaviour, AUDIT-C, and our secondary outcome measure, derived using graduated frequency questions, may be subject to self-report biases that may interact with the intervention. For example, the announcement of lower guidelines may lead drinkers to report lower alcohol consumption than they would have done otherwise. This problem is common to all survey-based studies of alcohol consumption,164 and our objective secondary outcome measures derived from alcohol taxation and hospital admission data are useful in this regard. Second, the UK Government and CMOs chose to announce the revised guidelines in January 2016. This hinders evaluation as December is typically a heavy drinking month and January a light drinking month. Although our models seek to control for seasonal trends, our time series remains relatively short and this control may be inadequate if seasonal trends vary substantially from year to year. Third, our descriptive analyses and our analyses of graduated frequency rely on questions that were included in the ATS for only 2 months prior to the announcement of revised guidelines. Although we are able to observe post-intervention trends in measures derived from these questions, our knowledge of how these differ from the pre-intervention period is limited. Fourth, a number of our survey questions have limitations. Self-reported measures are known to underestimate alcohol consumption,153,164 although we mitigate this by also including objective outcome measures derived from alcohol taxation and hospitalisation data. Our measures of exposure to the guidelines relate only to last-month exposure and are asked only to drinkers who are aware of guidelines and are able to provide a figure for the guideline threshold, whether correct or incorrect. Thus, we cannot talk about exposure of drinkers in general to guidelines but only about exposure of this select group within a defined time period. Fifth, our measures of influence on behaviour change are informed by theory and measures shown to predict tobacco consumption but have not been formally validated. Finally, although the change in guidelines for men was greater than for women, the primary analysis did not include a test for a sex-specific effect. This is because the full population analyses found no substantial or sustained effect of the guidelines and the inconsistent findings across outcome measures and analytical approaches suggest that any subgroup analyses would be unlikely to produce results leading to clear conclusions.